David Marr on two types of information-processing problems — LessWrong